SHAI

shai is a coding agent, your pair programming buddy that lives in the terminal. Written in rust with love <3

Features

• Interactive coding agent — Chat with shai in your terminal to write code, fix bugs, and get answers
• Headless mode — Pipe prompts directly into shai for scripting and automation
• HTTP server — Run shai as a service with OpenAI-compatible APIs and SSE streaming
• Shell assistant — Automatically suggests fixes when commands fail in your terminal
• Project context — Load project-specific information via AGENTS.md files
• MCP Support — Configure specialized agents with MCP and OAuth support
• Skills — Extend shai with composable, on-demand procedural instructions
• Multiple LLM providers — Works with OVHcloud, OpenAI, Anthropic, Mistral, Ollama, OpenRouter

Installation

Latest stable release

curl -fsSL https://raw.githubusercontent.com/ovh/shai/main/install.sh | sh

Nightly version

curl -fsSL https://raw.githubusercontent.com/ovh/shai/main/install.sh | SHAI_RELEASE=unstable sh

The shai binary will be installed in $HOME/.local/bin

Quick Start

By default shai uses OVHcloud as an anonymous user meaning you will be rate limited! If you want to sign in with your account or select another provider, run:

shai auth

Once you have a provider set up, you can run shai:

shai

Usage

Interactive Mode

Simply run shai to start the interactive coding agent:

shai                          # Start TUI
shai -a myagent              # Start TUI with a custom agent
shai session                 # Start TUI with session picker
shai session <id>            # Start TUI restoring a specific session
shai session latest          # Start TUI restoring the most recent session
shai -r <session-id>        # Restore a session by ID
shai --latest                # Restore the most recent session

Headless Mode

Run shai without the TUI for scripting and automation:

shai -p "make me a hello world in main.py"
echo "make me a hello world in main.py" | shai

Chain multiple shai calls together:

echo "make me a hello world in main.py" | shai --trace | shai -p "now run it!"

Interactive Pipe Mode

Pipe content into shai and continue the conversation in the TUI:

echo "analyze this code" | shai -i

HTTP Server Mode

Run shai as an HTTP service with SSE streaming support:

shai serve --port 3000

Available API endpoints:

• POST /v1/chat/completions — OpenAI Chat Completions API (ephemeral mode)
• POST /v1/responses — OpenAI Responses API (stateful/stateless)
• GET /v1/responses/{id} — Get response by ID
• POST /v1/responses/{id}/cancel — Cancel a response
• POST /v1/multimodal — Simple multimodal API (streaming)
• POST /v1/multimodal/{session_id} — Simple multimodal API (with session)

Shell Assistant

shai can also act as a shell assistant — when a command fails, it will propose a fix. This works by injecting command hooks while monitoring your terminal output.

shai on      # Start monitoring
shai off     # Stop monitoring
shai status   # Check if shai is active

Configuration

Compatible OVHcloud endpoints

OVHcloud AI Endpoints provides OpenAI-compatible LLM endpoints for using shai with tools. Follow the official getting started guide to create a Public Cloud project and then generate an API key from the AI Endpoints section of the control panel.

Authenticating with an API key raises the rate limit from 2 requests per minute (anonymous, per IP and model) to 400 requests per minute (per project and model).

Once your key is set in shai, pick a model from the AI Endpoints catalog:

• models with a function calling feature work best with shai;
• for models without it, force structured output with the /set so option.

Project Context File

Shai supports the AGENTS.md convention — a standard markdown file at the root of your project containing build steps, code style guidelines, testing instructions, and other context for AI coding agents.

Shai loads context from two files (in priority order):

1. AGENTS.md — the canonical project context file (recommended).
2. SHAI.md — legacy override/supplement (deprecated, use AGENTS.md instead).

For nested projects in a monorepo, place an AGENTS.md in each subdirectory — the closest one to the edited file takes precedence.

Custom Agents

Instead of a single global configuration, you can create custom agent configurations with specific providers, tools, and system prompts.

shai agent           # Open agent picker in TUI
shai agent myagent   # Launch TUI with a specific agent
shai list agent      # List available agents

Agent configs live in ~/.config/shai/agents/. See Agent Configuration for the full schema.

Skills

Skills are composable, on-demand procedural instructions that extend shai's capabilities:

.shai/skills/
└── my-skill/
    └── SKILL.md

They are discovered from .shai/skills/ (project), ~/.config/shai/skills/ (global), and built-ins embedded in the binary (e.g. memory, the persistent-memory protocol). A same-named skill in a higher tier shadows the lower one, so you can override built-ins with your own version. See Skills documentation for details.

TUI Slash Commands

Command	Description
/auth	Configure AI provider
/agent	Switch agent (keeps conversation context)
/restore [index|id]	Restore a previous session
/latest	Restore the most recent session
/temp <float>	Set sampling temperature
/tc <method>	Set tool call method (auto, fc, fc2, so)
/theme [dark|light|toggle]	Set or toggle theme
/tokens	Display token usage
/tools	List all registered tools
/skills	List available skills
/mcp	List MCP servers and connection status
/regenerate	Regenerate the last response
/exit	Exit the TUI

See TUI Configuration for keyboard shortcuts and customization.

Development

git clone git@github.com:ovh/shai.git
cd shai
cargo build --release

Benchmarks

Shai includes a measurement harness that runs the agent against scripted tasks. See Benchmarks documentation for details.